E. Bulog et al., "Deep Learning Assisted Product Grouping for Shoe Size Recommendation", Proc. of 3DBODY.TECH 2022 - 13th Int. Conf. and Exh. on 3D Body Scanning and Processing Technologies, Lugano, Switzerland, 25-26 Oct. 2022, #63, https://doi.org/10.15221/22.63.

Title:

Deep Learning Assisted Product Grouping for Shoe Size Recommendation

Authors:

Eugene BULOG 1, Calli BATES 1, Naomi NORTH 2, Tsuyoshi IETA 2, Bo LI 1

1 ZOZO New Zealand Ltd., Auckland, New Zealand;
2 ZOZO., Tokyo, Japan

Abstract:

Shoe size recommendation tailored to specific products and users is a complex problem influenced by many factors. These include not only user-based attributes such as individual 3D foot shape and preferences, but also the sizing properties unique to each model of shoe. Large scale data collection and grouping of shoes based on the way they fit users is a crucial step towards being able to recommend to a user their perfect size in a specific item of footwear, down to the brand and product level.
This work presents a scalable and robust platform to facilitate AI-assisted grouping of footwear SKUs, allowing businesses to rapidly aggregate shoe products into groups containing similar items across multiple retailers with the exact same fitting properties, which can then be used to train a family of bespoke size recommendation models. These recommendation models use a combination of learned properties of each shoe and 3D foot scan data from users to predict a personalized ideal fitting size.
The platform leverages "human-in-the-loop" machine learning, by presenting highly accurate grouping predictions (generated by a deep learning triplet loss model) to human supervisors for quick confirmation. This provides a much faster alternative to humans combing an enormous list of products and manually cross checking each product against all existing groups.
Use of this platform has greatly accelerated the ability of our shoe size recommendation product (ZOZOMAT) to support new models of shoes - by automating the most time-intensive and error-prone aspect of grouping shoes for training and prediction. This results in more accurate and granular shoe size recommendations for users, and lower customer return rates in purchased shoes.

Keywords:

shoe size prediction, recommendation, deep learning, metric learning, triplet loss, human-in-the-loop learning, 3d foot scanning, automation, data collection, data cleaning

Details:

Full paper: 2263bulog.pdf
Proceedings: 3DBODY.TECH 2022, 25-26 Oct. 2022, Lugano, Switzerland
Paper id#: 63
DOI: 10.15221/22.63
Presentation video: 3DBodyTech2022_63_bulog.mp4

Copyright notice

© Hometrica Consulting - Dr. Nicola D'Apuzzo, Switzerland, hometrica.ch.
Reproduction of the proceedings or any parts thereof (excluding short quotations for the use in the preparation of reviews and technical and scientific papers) may be made only after obtaining the specific approval of the publisher. The papers appearing in the proceedings reflect the author's opinions. Their inclusion in these publications does not necessary constitute endorsement by the editor or by the publisher. Authors retain all rights to individual papers.

Note: click the + on the top left of the page to open/close the menu.